We analyze how modeling international dependencies improves forecasts for the global economy based on a Bayesian GVAR with SSVS prior and stochastic volatility. To analyze the source of performance gains, we decompose the predictive joint density into its marginals and a copula term capturing the dependence structure across countries. The GVAR outperforms forecasts based on country-specific models. This performance is solely driven by superior predictions for the dependence structure across countries, whereas the GVAR does not yield better predictive marginal densities. The relative performance gains of the GVAR model are particularly pronounced during volatile periods and for emerging economies.JEL classification: C53, E37, F47 Keywords: GVAR, global economy, forecast evaluation, log score, copula * Corresponding author: Florian Huber, Oesterreichische Nationalbank (OeNB), Phone: +43-1-404 20-5208. E-mail: florian.huber@oenb.at. We would like to thank Hans Manner and all participants of the AWI seminar as well as the OeNB seminar for helpful comments on earlier drafts of this paper. The opinions expressed in this paper are those of the authors and do not necessarily reflect the official viewpoint of the OeNB or of the Eurosystem.